Improving Sentiment Classification using Supervised Learning

Sentiment classification is one of the techniques of sentiment analysis. It refers to the computational techniques for classifying whether the sentiments of text are positive or negative. SentiWordNet is a popularly used lexicon for extraction of sentiment words from documents. Many existing opinion mining or sentiment classification models works on positive and negative polarities while ignoring the objective words. If More than 90 percent of the words in SentiWordNet are objective words, Sentiment classification may be affected due to this major portion of objective words. This article proposes a MovieSentiNet framework, which will reassign a proper sentiment value & tendency to objective words aiming at improving sentiment classification. The sentiment values to objective words are assigned based on three steps. (a) Extracting sentiments on the words from SentiWordNet, (b) Calculating relevance of objective words based on frequency of objective words in positively & negatively tagged documents. (c) Then performing sentiment classification using SVM. The experimental results on a 1000 positively & 1000 negatively tagged documents demonstrate the effectiveness of proposed approach. Original objective words that are reassigned with sentimental orientation have contributed in the improvement of sentiment classification.